检索结果-内蒙古大学图书馆

Continual Action Assessment via Task-Consistent Score-Discriminative Feature Distribution Modeling

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Li, Yuan-Ming Zeng, Ling-An Meng, Jing-Ke Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou510006 China School of Artificial Intelligence Sun Yat-sen University Zhuhai519082 China School of Computer Science and Engineering Sun Yat-sen University Guangdong Guangzhou510275 China The Guangdong Key Laboratory of Information Security Technology Sun YatsenUniversity Guangdong Guangzhou510275 China KeyLaboratory of Machine Intelligence and Advanced Computing Sun YatsenUniversity Ministry of Education Guangdong Guangzhou510275 China

Action Quality Assessment (AQA) is a task that tries to answer how well an action is carried out. While remarkable progress has been achieved, existing works on AQA assume that all the training data are visible for training at one time, but do not enable continual learning on assessing new technical actions. In this work, we address such a Continual Learning problem in AQA (Continual-AQA), which urges a unified model to learn AQA tasks sequentially without forgetting. Our idea for modeling Continual-AQA is to sequentially learn a task-consistent score-discriminative feature distribution, in which the latent features express a strong correlation with the score labels regardless of the task or action types. From this perspective, we aim to mitigate the forgetting in Continual-AQA from two aspects. Firstly, to fuse the features of new and previous data into a score-discriminative distribution, a novel Feature-Score Correlation-Aware Rehearsal is proposed to store and reuse data from previous tasks with limited memory size. Secondly, an Action General-Specific Graph is developed to learn and decouple the action-general and action-specific knowledge so that the task-consistent score-discriminative features can be better extracted across various tasks. Extensive experiments are conducted to evaluate the contributions of proposed components. The comparisons with the existing continual learning methods additionally verify the effectiveness and versatility of our approach. Data and code are available at https://***/iSEE-laboratory/Continual-AQA. Copyright © 2023, The Authors. All rights reserved.

关键词： Learning systems

Halftone Image Steganography Based on Reassigned Distortion Measurement 7th

学校读者我要写书评

暂无评论

Halftone Image Steganography Based on Reassigned Distortion...

7th International Conference on Artificial intelligence and Security, ICAIS 2021

作者： Xu, Wenbo Liu, Wanteng Lin, Cong Wang, Ke Wang, Wenbin Lu, Wei School of Computer Science and Engineering Guangdong Key Laboratory of Information Security Technology Ministry of Education Key Laboratory of Machine Intelligence and Advanced Computing Sun Yat-sen University Guangzhou510006 China School of Statistics and Mathematics Guangdong University of Finance and Economics Guangzhou510320 China Guangdong Science and Technology Innovation Monitoring and Research Center Guangzhou510033 China

ISBN: (纸本)9783030786113

Most state-of-the-art halftone image data hiding methods aim to preserve good visual quality when embedding messages, while ignoring the statistical security. This paper proposed a halftone steganographic scheme that improves the visual quality and the statistical security of the anti-steganalysis. First, a general distortion measurement for halftone images based on human visual system (HVS) model is proposed. Utilizing the Least-Mean-Square (LMS) method, halftone images can be converted to grayscale images and the objective image quality assessment is applied to evaluate the distortion caused by flipping pixels. Different distortion measurements can be derived from different image quality assessments. Then, to further measure the embedding distortions, we combine these distortion measurements to construct a reassigned distortion measurement based on the controversial pixels prior (CPP) rule. Finally, syndrome-trellis code (STC) is employed to minimize the number of flipping pixels. Experimental results have shown that the proposed steganographic scheme achieves strong statistical security with high capacity and visual quality. © 2021, Springer Nature Switzerland AG.

关键词： Image quality

Language-specific Characteristic Assistance for Code-switching Speech Recognition

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Song, Tongtong Xu, Qiang Ge, Meng Wang, Longbiao Shi, Hao Lv, Yongjie Lin, Yuqin Dang, Jianwu Tianjin Key Laboratory of Cognitive Computing and Application College of Intelligence and Computing Tianjin University Tianjin China Department of Electrical and Computer Engineering National University of Singapore Singapore Singapore Graduate School of Informatics Kyoto University Sakyo-ku Kyoto Japan Japan Advanced Institute of Science and Technology Ishikawa Japan

Dual-encoder structure successfully utilizes two language-specific encoders (LSEs) for code-switching speech recognition. Because LSEs are initialized by two pre-trained language-specific models (LSMs), the dual-encoder structure can exploit sufficient monolingual data and capture the individual language attributes. However, most existing methods have no language constraints on LSEs and underutilize language-specific knowledge of LSMs. In this paper, we propose a language-specific characteristic assistance (LSCA) method to mitigate the above problems. Specifically, during training, we introduce two language-specific losses as language constraints and generate corresponding language-specific targets for them. During decoding, we take the decoding abilities of LSMs into account by combining the output probabilities of two LSMs and the mixture model to obtain the final predictions. Experiments show that either the training or decoding method of LSCA can improve the model's performance. Furthermore, the best result can obtain up to 15.4% relative error reduction on the code-switching test set by combining the training and decoding methods of LSCA. Moreover, the system can process code-switching speech recognition tasks well without extra shared parameters or even retraining based on two pre-trained LSMs by using our method. Copyright © 2022, The Authors. All rights reserved.

关键词： Speech recognition

A Novel Deep Learning Approach Featuring Graph-Based Algorithm for Cell Segmentation and Tracking

学校读者我要写书评

暂无评论

A Novel Deep Learning Approach Featuring Graph-Based Algorit...

IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

作者： Keliang Zhao Jovial Niyogisubizo Linxia Xiao Yi Pan Dongqing Wei Didi Rosiyadi Yanjie Wei Shenzhen Key Laboratory of Intelligent Bioinformatics and Center for High Performance Computing Shenzhen Institute of Advanced Technology Chinese Academy of Sciences Shenzhen China College of Computer Science and Control Engineering Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Shenzhen China State Key Laboratory of Microbial Metabolism and School of Life Sciences and Biotechnology Shanghai Jiao Tong University Shanghai China Research Center for Artificial Intelligence and Cybersecurity National Research and Innovation Agency Bandung Indonesia

The precise segmentation and tracking of cells in microscopy image sequences play a pivotal role in biomedical research, facilitating the study of tissue, organ, and organism development. However, manual segmentation and tracking of cells is time-consuming and often require professional experiences. Besides, segmenting cells in the images with a low signal-to-noise ratio remains difficult. While deep learning (DL) has become a common method for cell segmentation, few DL- based methods address concurrent cell segmentation and tracking. In this paper, we propose a novel DL approach featuring graph-based tracking for cell segmentation and tracking in microscopy images. We combine Deeplabv3+ for semantic segmentation and ResNet50 for enhanced feature extraction, enabling comprehensive cell detection and instance segmentation. Post-processing, involving non-maxima suppression and outlier detection, refines predictions and produces final segmentation. The tracking method is based on the relative position of graph nodes to track segmented cells, encompassing cell division and apoptosis. We conduct our experiments on the induced pluripotent stem (iPS) cell datasets, and the results show that the segmentation and tracking performance of our method yields superior performance compared to the benchmark models. More specifically, our approach achieved DET values of 0.955 and 0.913, TRA values of 0.951 and 0.906, and SEG values of 0.690 and 0.665 on two iPS dataset videos, respectively. Additionally, the performance assessment encompasses four real microscopy datasets from the Cell Tracking Challenge (CTC). Our method greatly reduces the cost of manual labeling and labor-intensive costs.

关键词：

Unveiling Hidden Vulnerabilities in Digital Human Generation via Adversarial Attacks

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Li, Zhiying Jin, Yeying Shen, Fan Liu, Zhi Chen, Weibin Zhang, Pengju Zhang, Xiaomei Chen, Boyu Shen, Michael Wu, Kejian Fan, Zhaoxin Dong, Jin College of Cyber Security Jinan University China Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing School of Artificial Intelligence Beihang University China Hangzhou International Innovation Institute Beihang University China Department of Electrical and Computer Engineering National University of Singapore Singapore Department of Electrical and Computer Science University of Pittsburgh United States State Key Laboratory of Multimodal Artificial Intelligence Institute of Automation Chinese Academy of Science China Institute of Automation Chinese Academy of Science China University College London United Kingdom Mingdu Tech China Xreal Beijing Academy of Blockchain and Edge Computing China

Expressive human pose and shape estimation (EHPS) is crucial for digital human generation, especially in applications like live streaming. While existing research primarily focuses on reducing estimation errors, it largely neglects robustness and security aspects, leaving these systems vulnerable to adversarial attacks. To address this significant challenge, we propose the Tangible Attack (TBA), a novel framework designed to generate adversarial examples capable of effectively compromising any digital human generation model. Our approach introduces a Dual Heterogeneous Noise Generator (DHNG), which leverages Variational Autoencoders (VAE) and ControlNet to produce diverse, targeted noise tailored to the original image features. Additionally, we design a custom adversarial loss function to optimize the noise, ensuring both high controllability and potent disruption. By iteratively refining the adversarial sample through multi-gradient signals from both the noise and the state-of-the-art EHPS model, TBA substantially improves the effectiveness of adversarial attacks. Extensive experiments demonstrate TBA’s superiority, achieving a remarkable 41.0% increase in estimation error, with an average improvement of approximately 17.0%. These findings expose significant security vulnerabilities in current EHPS models and highlight the need for stronger defenses in digital human generation systems. Copyright © 2025, The Authors. All rights reserved.

关键词： Man machine systems

Cross Languages One-Versus-All Speech Emotion Classifier 1

学校读者我要写书评

暂无评论

2nd International Conference on Neural computing for advanced Applications, NCAA 2021

作者： Liu, Xiangrui Bin, Junchi Li, Huakang Key Laboratory of Urban Land Resources Monitoring and Simulation Ministry of Natural Resources Shenzhen China School of Engineering University of British Columbia Kelowna Canada School of Artificial Intelligence and Advanced Computing Xi’an Jiaotong-Liverpool University Suzhou China Suzhou Privacy Technology Co. Ltd. Suzhou China

ISBN: (数字)9789811651885

ISBN: (纸本)9789811651878

Speech emotion recognition (SER) is a task that cannot be accomplished solely depending on linguistic models due to the presence of figures of speech. For a more accurate prediction of emotions, researchers adopted acoustic modelling. The complexity of SER can be attributed to a variety of acoustic features, the similarities among certain emotions, etc. In this paper, we proposed a framework named Cross Languages One-Versus-All Speech Emotion Classifier (CLOVASEC) that identifies speeches’ emotions for both Chinese and English. Acoustic features were preprocessed by Synthetic Minority Oversampling Technique (SMOTE) to diminish the impact of an imbalanced dataset then by Principal component analysis (PCA) to reduce the dimension. The features were fed into a classifier that was made up of eight sub-classifiers and each sub-classifier was tasked to differentiate one class from the other seven classes. The framework outperformed regular classifiers significantly on The Chinese Natural Audio-Visual Emotion Database (CHEAVD) and an English dataset from Deng. © 2021, Springer Nature Singapore Pte Ltd.

关键词： Principal component analysis

Simulations and Experiments of Time-Varying Takagi Factorization via Zhang Neural Network 5

学校读者我要写书评

暂无评论

Simulations and Experiments of Time-Varying Takagi Factoriza...

5th IEEE advanced Information Technology, Electronic and Automation Control Conference, IAEAC 2021

作者： Tang, Zanyu Zeng, Tianyu Li, Zhenyu Chen, Jianrong Zhang, Yunong Sun Yat-sen University School of Computer Science and Engineering Guangzhou510006 China Guangdong Key Laboratory of Modern Control Technology Guangzhou510070 China College of Information Science and Engineering Jishou University Jishou416000 China School of Public Health Administration Youjiang Medical University for Nationalities Baise533000 China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou510006 China

The problem of Takagi factorization (TF) for complex time-varying symmetric matrix is proposed in this paper. To solve this difficult and challenging problem, a continuous-time TF (CTTF) model of derivative equation system form is proposed via Zhang neural network (ZNN) method. For the sake of digital hardware implementation, we also build up the corresponding discrete-time (DT) ZNN solution models with different precisions of the derivative equation system discretization. Specifically, an eight-instant DTTF (EIDTTF) model are proposed. We also comparatively use two other multi-instant (e.g., four-instant and six-instant, i.e., FIDTTF and SIDTTF, respectively) models by discretizing the CTTF model. The effective solution step-size intervals of the proposed DTTF models are further discussed lastly. The results of simulations and experiments substantiate the correctness, effectiveness and superiority of the proposed CTTF and DTTF models. © 2021 IEEE.

关键词： Complex networks

Vision Transformer for Contrastive Clustering

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Ling, Hua-Bao Zhu, Bowen Huang, Dong Chen, Ding-Hua Wang, Chang-Dong Lai, Jian-Huang The College of Mathematics and Informatics South China Agricultural University Guangzhou China The School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangdong Key Laboratory of Information Security Technology Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education China

Vision Transformer (ViT) has shown its advantages over the convolutional neural network (CNN) with its ability to capture global long-range dependencies for visual representation learning. Besides ViT, contrastive learning is another popular research topic recently. While previous contrastive learning works are mostly based on CNNs, some recent studies have attempted to combine ViT and contrastive learning for enhanced self-supervised learning. Despite the considerable progress, these combinations of ViT and contrastive learning mostly focus on the instance-level contrastiveness, which often overlook the global contrastiveness and also lack the ability to directly learn the clustering result (e.g., for images). In view of this, this paper presents a novel deep clustering approach termed Vision Transformer for Contrastive Clustering (VTCC), which for the first time, to our knowledge, unifies the Transformer and the contrastive learning for the image clustering task. Specifically, with two random augmentations performed on each image, we utilize a ViT encoder with two weight-sharing views as the backbone. To remedy the potential instability of the ViT, we incorporate a convolutional stem to split each augmented sample into a sequence of patches, which uses multiple stacked small convolutions instead of a big convolution in the patch projection layer. By learning the feature representations for the sequences of patches via the backbone, an instance projector and a cluster projector are further utilized to perform the instance-level contrastive learning and the global clustering structure learning, respectively. Experiments on eight image datasets demonstrate the stability (during the training-from-scratch) and the superiority (in clustering performance) of our VTCC approach over the state-of-the-art. © 2022, CC BY.

关键词： Convolution

Strongly Augmented Contrastive Clustering

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Deng, Xiaozhi Huang, Dong Chen, Ding-Hua Wang, Chang-Dong Lai, Jian-Huang The College of Mathematics and Informatics South China Agricultural University Guangzhou China The School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangdong Key Laboratory of Information Security Technology Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education China

Deep clustering has attracted increasing attention in recent years due to its capability of joint representation learning and clustering via deep neural networks. In its latest developments, the contrastive learning has emerged as an effective technique to substantially enhance the deep clustering performance. However, the existing contrastive learning based deep clustering algorithms mostly focus on some carefully-designed augmentations (often with limited transformations to preserve the structure), referred to as weak augmentations, but cannot go beyond the weak augmentations to explore the more opportunities in stronger augmentations (with more aggressive transformations or even severe distortions). In this paper, we present an endto-end deep clustering approach termed Strongly Augmented Contrastive Clustering (SACC), which extends the conventional two-augmentation-view paradigm to multiple views and jointly leverages strong and weak augmentations for strengthened deep clustering. Particularly, we utilize a backbone network with triply-shared weights, where a strongly augmented view and two weakly augmented views are incorporated. Based on the representations produced by the backbone, the weak-weak view pair and the strong-weak view pairs are simultaneously exploited for the instance-level contrastive learning (via an instance projector) and the cluster-level contrastive learning (via a cluster projector), which, together with the backbone, can be jointly optimized in a purely unsupervised manner. Experimental results on five challenging image datasets have shown the superiority of our SACC approach over the state-of-the-art. The code is available at https://***/dengxiaozhi/SACC. © 2022, CC BY.

关键词： Deep neural networks